CABRA: Clustering algorithm based on regular arrangement

Abstract

Clustering is an unsupervised learning technique for organizing complex datasets into coherent groups. A novel clustering algorithm is presented, with a simple grouping concept depending on only one hyperparameter, which makes it suitable for further extensions to any topology and space. It is compared to state-of-the-art algorithms, overall achieving a better performance independently on the structure and complexity of the data, making the proposed algorithm a valuable tool for real applications such as market segmentation, sentiment analysis and anomaly detection.